- Ensure Malmo Mod for Minecraft is set
- Run Minecraft
- Collected trajectories saved in
.tgz
file in folderhuman_trajectories
. - Install gym_minigrid with dependencies with
setup.py
.
cd code
pip install -e .
Change the xmlfile name in FileWorldGenerator
cd code
python3 minecraft_to_minigrid_debug.py --xmlfile <path_to_xml>
Play the game to record the observed map and directional frontiers at each timestep. TODO: Record over some time interval like 5 steps?
numpy array saved with map
which is env_size x envsize, agent_dir
which is continuous yaw, subset_frontiers_mask
which is a boolean list of frontier, where the frontiers are calculated from top left to bottom right. Refer get_absolute_frontier_coordinates
in minecraft_to_minigrid_debug.py
We can also save the 3D tensor with channels as map
and subset_frontiers
.
3D tensor of shape env_size, env_size, num_channels where the channels contain the following : index 0 : trajectory map of the agent in absolute coordinates index 1 : subset of frontiers selected
Note: currently frontiers are selected based on cardinal directional input. To be extended to natural language input.